Nvidia’s 1,200% three-year surge has left it priced for perfection; Broadcom and AMD are now stepping into AI segments with faster growth and lower valuations, giving late-cycle investors a cleaner risk-reward setup.
The Aftermath of a 1,200% Rocket Ride
Nvidia’s revenue has jumped nearly ten-fold since 2022, pushing its market cap past $3 trillion and making it the world’s largest public company. The catalyst was an early lock on large language model training via its CUDA software stack and a ~90% share of the AI-GPU market. That dominance is now baked into the price: shares trade at 36× forward sales, leaving minimal room for disappointment.
Wall Street’s next questions are pragmatic: where does the incremental dollar go when Nvidia’s share can only fall, and who benefits as the AI build-out pivots from training to inference and custom silicon?
Broadcom: The ASIC Kingmaker
Enter Broadcom. Instead of selling one-size-fits-all GPUs, the company licenses its IP blocks and manages tape-outs for hyperscalers that want application-specific integrated circuits (ASICs). These chips sacrifice flexibility for brute efficiency—exactly what energy-hungry inference clusters require.
- Alphabet’s TPU lineage—now in its sixth generation—was built on Broadcom IP and remains one of the highest-volume AI accelerators outside Nvidia.
- OpenAI has engaged Broadcom to co-design its first in-house chip, targeting mass deployment before 2027.
- Apple is reportedly tapping Broadcom for an AI ASIC aimed at on-device and edge workloads.
The math is eye-catching: Broadcom’s AI-related revenue was roughly $20 billion in fiscal 2025. Citigroup projects that line surpassing $50 billion in fiscal 2026 and $100 billion by fiscal 2027—compound annual growth above 65%. At a current 14× forward sales multiple, the stock offers exposure to that surge at less than half Nvidia’s valuation.
AMD: Inference Market Share Grab
While Nvidia owns training, AMD is attacking the larger, still-fragmented inference arena where software portability matters less and total cost of ownership rules.
- Oracle will deploy 50,000 AMD MI300X GPUs in 2026, explicitly for inference workloads.
- OpenAI took a 10% equity stake and committed to 6 GW of AMD silicon—an order worth an estimated $20–25 billion over five years.
- Microsoft has ported CUDA libraries to AMD’s ROCm stack, removing a key adoption barrier.
Combined with its incumbent lead in server CPUs, AMD guidance calls for >60% data-center revenue CAGR through 2028. Even if GPUs capture only a mid-teens share of total AI accelerators, the revenue uplift could double AMD’s top line without requiring another hero product cycle.
Valuation Check: Where the Margin of Safety Lives
Nvidia at 36× sales needs perfection; Broadcom at 14× and AMD at 6× sales need merely execution. Both challengers carry net-cash balance sheets, buy back stock aggressively, and trade at forward PEG ratios below 1.2—levels Nvidia hasn’t seen since 2015.
What Could Go Wrong
- A sudden GPU inventory glut could pressure all chip prices.
- Custom-ASIC roll-outs are notoriously tape-out sensitive; a six-month delay would push revenue recognition into 2027.
- Geopolitical restrictions on TSMC production remain an industry-wide chokepoint.
Portfolio Playbook
Investors who missed Nvidia’s vertical take-off now face a classic late-cycle dilemma: chase a $3 trillion name or rotate into derivative upside. A barbell approach—core allocation to Broadcom’s ASIC story and a growth sleeve in AMD’s inference pivot—captures the next wave of AI cap-ex with asymmetric risk-reward and without paying Nvidia’s scarcity premium.
Stay ahead of the silicon cycle with instant, data-driven analysis—read more breaking financial insights at onlytrustedinfo.com.
